Boise State University

ScholarWorks
Center for Geophysical Investigation of the Shallow
CGISS Publications and Presentations
Subsurface (CGISS)

12-21-2007

Tomography of the Darcy Velocity from Self-

Potential Measurements
A. Jardani
Colorado School of Mines

A. Revil
Colorado School of Mines

A. Bolève
Institut National des Sciences de l'Univers

A. Crespy
Institut National des Sciences de l'Univers

J.-P. Dupont
University of Rouen

See next page for additional authors

Copyright 2007 by the American Geophysical Union. DOI: 10.1029/2007GL031907

Authors
A. Jardani, A. Revil, A. Bolève, A. Crespy, J.-P. Dupont, Warren Barrash, and Bwalya Malama

This article is available at ScholarWorks: http://scholarworks.boisestate.edu/cgiss_facpubs/30

 GEOPHYSICAL RESEARCH LETTERS, VOL. 34, L24403, doi:10.1029/2007GL031907, 2007
Click
Here
for
Full
Article
Tomography of the Darcy velocity from self-potential measurements
A. Jardani,1,2 A. Revil,1,3 A. Bolève,3,4 A. Crespy,3 J.-P. Dupont,5 W. Barrash,6
and B. Malama6
Received 1 September 2007; revised 31 October 2007; accepted 8 November 2007; published 21 December 2007.
[1] An algorithm is developed to interpret self-potential
(SP) data in terms of distribution of Darcy velocity of the
ground water. The model is based on the proportionality
existing between the streaming current density and the
Darcy velocity. Because the inverse problem of current
density determination from SP data is underdetermined, we
use Tikhonov regularization with a smoothness constraint
based on the differential Laplacian operator and a prior
model. The regularization parameter is determined by the
L-shape method. The distribution of the Darcy velocity
depends on the localization and number of non-polarizing
electrodes and information relative to the distribution of
the electrical resistivity of the ground. A priori hydraulic
information can be introduced in the inverse problem. This
approach is tested on two synthetic cases and on real SP
data resulting from infiltration of water from a ditch.
Citation: Jardani, A., A. Revil, A. Bolève, A. Crespy, J.-P.
Dupont, W. Barrash, and B. Malama (2007), Tomography of the
Darcy velocity from self-potential measurements, Geophys. Res.
Lett., 34, L24403, doi:10.1029/2007GL031907.
1. Introduction
[2] Geophysical methods such as ground-penetrating
radar, DC electrical resistivity tomography, electromagnetic
methods, induced polarization, seismic, and nuclear mag-
netic resonance imaging are sensitive to various hydraulic
parameters of porous and fractured materials through the
detection of changes in soil physical properties over time.
The self-potential (SP) method is the only geophysical
method that is directly sensitive to the flow of the ground
water [e.g., Revil et al., 2005]. Non-polarizing electrodes
can therefore be considered as a non-intrusive flow sensor.
Jardani et al. [2006] inverted recently SP data to reconstruct
a boundary between two formations characterized by a net
divergence of the streaming current density with the assump-
tion that the vadose zone is more resistive than the aquifer.
Minsley et al. [2007] proposed an algorithm to invert SP data
in terms of the divergence of the streaming current density.
1
Department of Geophysics, Colorado School of Mines, Golden,
Colorado, USA.
2
Bureau d’Etudes Action Locale et Internationale pour la Solidarité et
l’Environnement, Saint-Jacques-sur-Darnétal, France.
3
UMR 5559, Laboratoire de Géophysique Interne et Tectonophysique,
CNRS, Institut National des Sciences de l’Univers, Chambéry, France.
4
Also at SOBESOL, Savoie Technolac, Le Bourget du Lac Cedex,
France.
5
UMR 6143, CNRS, Département de Géologie, University of Rouen,
Rouen, France.
6
Center for Geophysical Investigation of the Shallow Subsurface and
Department of Geosciences, Boise State University, Boise, Idaho, USA.
Copyright 2007 by the American Geophysical Union.
0094-8276/07/2007GL031907$05.00
L24403
[3] As far as we know, the inversion of the streaming
current density has never been attempted. The determination
of this distribution would be very useful to determine the
distribution of the Darcy velocity using the pore-scale
electrokinetic model validated recently by Bolève et al.
[2007a]. We propose below an algorithm to map ground
water flow non-intrusively from SP data. This offers the
possibility to monitor, in real time, ground water flow in the
subsurface of the Earth. In this letter, we focus only on
simple cases. We feel, however, applications to more
complex systems will require the combination of SP inver-
sion with additional geophysical methods and with in situ
measurements of groundwater heads and/or temperature.
2. Forward Modeling
[4] Self-potential signals of electrokinetic nature are due
to the drag of the excess of electrical charge contained in the
pore water and resulting from the existence of the electrical
diffuse layer at the pore water/mineral interface. In an
isotropic but possibly heterogeneous medium, the total
current density is given by [Revil et al., 2007; Bolève et
al., 2007b]
qs
j ¼ sðqÞr8 þ QV u; ð1Þ
q
where qs is the porosity, q is the water content, j is the
electrical current density (in A m2), u is the Darcy velocity
(in m s1), 8 is the SP (in V), s is the electrical conductivity of
the porous material (in S m1) [see Revil et al., 1998], and QV
is the excess charge (of the diffuse layer) of the pore water per
unit pore volume (in C m3), which depends mainly on the
permeability of the porous material (Figure 1). The continuity
equation for the electrical charge is r  j = 0. The Richards
equation is u = (krKs)r(H + z) [Richards, 1931], where kr is
the relative permeability, Ks the hydraulic conductivity at
saturation, and H = dp/rfg is the change in hydraulic head at a
given elevation (above or below the hydrostatic initial
distribution H0), dp is the change of pressure relative to the
hydrostatic level, rf is the pore fluid density (in kg m3), and g
is the acceleration of the gravity (in m s2). At saturation, the
streaming potential coupling coefficient is defined by C =
(@8/@H)j = 0 = QVK/s. The measurement of C can be used to
determine the values of QV [Bolève et al., 2007b]. In Figure 1,
we have reported the value of QV for different measurement
of C. We observe that for a variety of rocks and ionic strengths
of the pore water, QV depends mainly on the permeability of
the porous rock.
[5] The hydraulic problem can be solved using the
continuity equation,
@H
½Ce þ Se SS þ r  u ¼ 0; ð2Þ
@t
1 of 6
L24403 JARDANI ET AL.: TOMOGRAPHY OF GROUND WATER FLOW L24403

steady-state or transient conditions (forward problem). The

forward problem was validated with several data sets. In
section 3, we invert the distribution of the current density jS =
QV(qs/q)u from SP data.

3. Inverse Modeling
[7] The relationship between the electrical current density
at point M and the measured SP signals at non-polarizing
electrode P can be written as

Z
8ð PÞ ¼ Kð P; M ÞjS ð M ÞdV ; ð6Þ
W

where jS is the source current density (in both saturated and

Figure 1. Dependence of the excess of charge per unit
pore volume with the permeability at saturation of the
porous materials. This data set includes various lithologies
and salinities (6  pH  8.5). The alluvium materials are
from the Boise State University test site. They are composed
of a mixture of sands and gravels.
where H (m) is the pressure head, z is the altitude above a
datum, Ce denotes the specific moisture capacity (in m1)
defined by Ce = @q/@H where q is the water content
(dimensionless), Se is the effective saturation, which is
related to the relative saturation of the water phase by Se =
(q  qr)/(qs  qr) where qr is the residual water content.
[6] With the van Genuchten parameterization, we con-
sider the soil to be saturated when the fluid pressure reaches
the atmospheric pressure (H = 0). The effective saturation,
the specific moisture capacity, and the relative permeability
are given by
8
m
>
< 1= 1 þ jaH jn ; H < 0
Se ¼ ð3Þ
>
:
1; H  0
unsaturated conditions) described in section 2 and K(P, M)
is the kernel connecting the SP data measured at a set of
non-polarizing electrodes P (with respect to a reference
electrode) and the source of current at point M in the
conducting ground. The kernel K depends on the number of
measurement stations at the ground surface, the number of
discretized elements in which the source current density is
going to be determined, and the resistivity distribution of
the medium. The inversion of the SP data follows a two-step
process. The first step is the inversion of the distribution of
the current density jS. The second step is the determination
of u using the distribution of jS and assuming values for the
excess charge density and the ratio (qs/q) for unsaturated
conditions.
[8] This SP inverse problem is a typical (vectorial)
potential field problem and the solution of such problem
is known to be ill-posed and non-unique. It is therefore
important to add additional constraints to reduce the space
of the solution. The criteria of data misfit and model
objective function place different and competing, require-
ments on the models. These objective functions are bal-
anced using Tikhonov regularization [Tikhonov and
Arsenin, 1977] through the definition of a global objective
function, y,
y ¼ kWd ðKm  8d Þk2 þ lkWm ðm  m0 Þk2 ; ð7Þ

8 am 1
 
1 m
< ðf  qr ÞSem 1  Sem ; H < 0
Ce ¼ 1  m ð4Þ where kAfk2 = f tAtAf (t is transpose), l is a regularization
:
0; H  0 parameter under the constraint that (0 < l < 1), K = (Kxij,
Kzij) is the kernel Nx2M matrix corresponding to the SP,
which can be measured by each component of a source at
8 h   i coordinates m = (jxi , jzi ) and where N is the number of SP
1 m 2
>
< S L 1  1  Sem
e ;H < 0 stations while M is the number of discretized cells
kr ¼ ð5Þ
>
: composing the ground, 2M represents the number of
1; H  0 elementary current sources to consider (one horizontal
component and one vertical component per cell for a 2D
respectively, and a, n, m = 1  1/n, and L are dimensionless problem), 8d is vector of N elements corresponding to
constants that characterize the porous material [van the SP data measured at the ground surface or in boreholes,
Genuchten, 1980]. Bolève et al. [2007b] used the commer- Wd = diag{1/e1,. . .,1/eN} is a square diagonal weighting
cial finite element software Comsol Multiphysics 3.3 to NxN matrix (elements along the diagonal of this matrix are
determine the SP distribution associated with ground water the reciprocals of the standard deviations ei of the data), Wm
flow in saturated and unsaturated conditions under either

2 of 6
L24403 JARDANI ET AL.: TOMOGRAPHY OF GROUND WATER FLOW L24403

Figure 2. Synthetic case for the 2D-infiltration from a ditch. (a) The Darcy velocity is modeled from the Richard
equation. (b) Distribution of the self-potential for the synthetic case. We assume that the «measurements» of the self-
potentials are performed at the location with the symbols (+) are located. (c) Distribution of the reconstructed Darcy
velocity (R2 = 0.98).

is a 2(M  2)  2M weighting matrix (e.g., the flatness
matrix or the differential Laplacian operator), m is the
vector of 2M model parameters (source current density), and
m0 is a reference model (i.e., prior distribution of the source
current density). Wm is given by Zhdanov [2002] and we
consider that the probability distribution of the SP
measurements is Gaussian [Linde et al., 2007].
[9] The solution of the problem corresponding to the
minimum of the cost function is [Hansen, 1998]:
The solution depends on the value of the regularization
parameter l and the a prior model m0. To determine the
value of l, Hansen [1998] proposed plotting the norm of
the regularized smoothing solutions versus the norm of the
residuals of the data misfit function. This dependence often
has an L-shaped form and the best regularization parameter
lies on the corner of the L-shape curve. If we use a null
distribution of prior information (m0 = 0), the previous
model is similar to a damped weighted linear least squares
or biased linear estimation problem. However, it is also
 

1 possible to estimate the a prior model by simulating the flow
m* ¼ KT WTd Wd K þ l WTm Wm
 


of the ground water assuming an homogeneous subsoil,
 KT WTd Wd 8d þ l WTm Wm m0 : ð8Þ using the appropriate boundary conditions, and finally
converting the seepage velocity in an a prior distribution of

3 of 6
L24403 JARDANI ET AL.: TOMOGRAPHY OF GROUND WATER FLOW L24403

Figure 3. Synthetic case for a vertical 2D-infiltration problem. (a) True distribution of the Darcy velocity (in m s1).
(b) Distribution of the resulting self-potentials, which is sampled at the top surface of the system every meter.
(c) Distribution of the reconstructed Darcy velocity (R2 = 0.95).

the current density using a constant value for QV. We will
use both approaches in the following examples.
4. Synthetic Cases
[10] The inverse model was set up in a Matlab routine. To
test this routine, we simulated the case of the flow of the
ground water from a ditch in a small thin tank with a length
of 2 m and a height of 0.5 m. The fictitious tank is assumed
to be filled with a porous material with constant properties
(ssat = 0.012 S m1, f = 0.33, Ks = 8  105 m s1, and
Csat = 3 mV m1). This yields a volumetric charge density
QV = 0.48 C m3 with hf = 1.14 103 Pa s. The flux is
imposed at the ditch (0.4 mm s1). The boundaries of
the tank are both impermeable and insulating (n.u = 0 and
n.j = 0). The Richards equation is solved with Comsol
Multiphysics 3.3 [Bolève et al., 2007b]. Figures 2a and 2b
show the distribution of the Darcy velocity and the
resulting SP field at t = 300 s. The material properties are
a = 1.54, n = 7.6, qr = 0, and L = 0.5. We assume that the SP
distribution is sampled at 28 non-polarizing electrodes
(indicated by the crosses on Figure 2a). Then, this SP data
are inverted to determine the distribution of the source
current density using a null distribution of prior information
(Figure 2c). Finally, we use the inverted distribution of
the current density to determine the Darcy velocity. The
distribution of the Darcy velocity is very similar to the
modeled ground water flow pattern. The magnitude of
the Darcy velocity is slightly smaller than the true Darcy
velocity of the model.
[11] A second synthetic case is shown in Figure 3. This
time, we model the vertical flow path due to an heteroge-
neity in the distribution of Ks. We have a uniform
background medium (Ks = 104 m s1, f = 0.33, ssat =

4 of 6
L24403 JARDANI ET AL.: TOMOGRAPHY OF GROUND WATER FLOW
0.012 S m1) and a less permeable layer (Ks = 107 m s1,
f = 0.40, ssat = 0.10 S m1) that contains a discontinuity.
The flux is imposed from the top surface of the system
(1 mm s1) and computations are performed in transient
conditions with Comsol Multiphysics 3.3. The results are
shown at t = 60 s (Figure 3). Then, the SP information
determined at the ground surface is used to retrieve the
position of the permeable pathway using a null distribution
of prior information. We determine the distribution of the
Darcy velocity (from its strength and the distribution of QV,
which is connected to the distribution of the permeability,
Figure 1). Again the inverted values have a smoother
distribution than the true Darcy velocity and a slightly
smaller magnitude, but the final results compare well with
the ‘‘true’’ model.
5. Application to an Infiltration Test
[12] We analyze now the infiltration experiment reported
by Suski et al. [2006] and carried out at the test site of
Roujan in the southern part of France. Eighteen piezometers
were installed to a depth of 4 m on one side of a ditch,
which is 0.8 m deep, 1.5 m wide, and 10 m long [Suski et
al., 2006]. The SP signals were monitored using a network
of 41 PMS9000-Pb/PbCl2 electrodes. Electrical resistivity
tomography (ERT) along a section perpendicular to the
ditch indicates that the resistivity of the soil was 20 Wm
except for the first 50 cm where the resistivity was
100 Wm. The piezometers show that the water table was
initially located at 2 m below the ground surface. During the
experiment, 14 m3 of fresh water were injected in the ditch.
Laboratory experiments yields C = 5.8 ± 1.1 mV m1.
The SP profile is can be found in the work by Suski et al.
[2006]. Because of the symmetry of the problem with an
axis of symmetry corresponding to the ditch, only one side
of the ditch is modeled.
[13] To perform the inverse problem, the a prior model
m0 is setup using the solution of the flow model at the time
at which the self-potential measurements were obtained
(170 minutes after the start of the infiltration). The result
of the inversion is shown on Figure 4. Note that in total, the
model uses 294 cells for its discretization but only the
64 cells close to the ditch are shown in Figure 4). A 2D
numerical simulation was performed with Comsol Multi-
physics 3.3 along a cross-section perpendicular to the ditch
(Figure 5). We use the full formulation including capillary
Figure 4. Reconstruction of the Darcy velocity from the
self-potential data of Suski et al. [2006] (R2 = 0.98) using an
a prior model.
5 of 6
L24403
Figure 5. Simulated SP distribution and the Darcy
velocity associated with the infiltration in the ditch of
Roujan using all the hydraulic information available from in
situ and petrophysical measurements [see Bolève et al.,
2007b].
effects and heterogeneity in the distribution of the electrical
resistivity [see Bolève et al., 2007b]. Inside the ditch, we
imposed a hydraulic head that varies over time according to
the measured water level [Suski et al., 2006]. The
hydrogeological model and the values of QV used to
perform the simulation is reported by Bolève et al. [2007b].
A snapshot of the SP and Darcy velocity distributions in the
course of the infiltration is shown in Figure 5. The
distribution of the Darcy velocities agrees reasonably well
with the result inverted from the SP data (Figure 4).
6. Conclusion
[14] We used a finite element code (Comsol Multiphysics
3.3) within an inverse model based on Tikhonov regulari-
zation (in a Matlab routine) to consider the semi-coupled
differential equations of fluid flow and electrical current
density. The inversion of SP data with a constraint of
smoothness allows the determination of the distribution of
ground water flow velocity in the subsurface of the Earth for
moderately heterogeneous media.
[15] Acknowledgments. This work is supported by ANR and ECCO
programs ‘‘ERINOH’’, ‘‘POLARIS’’, INRA-ECCO, and ALISE ENVI-
RONNEMENT. We thank T. Young for his support.
References
Bolève, A., A. Crespy, A. Revil, F. Janod, and J. L. Mattiuzzo (2007a),
Streaming potentials of granular media: Influence of the Dukhin and
Reynolds numbers, J. Geophys. Res., 112, B08204, doi:10.1029/
2006JB004673.
Bolève, A., A. Revil, F. Janod, J. L. Mattiuzzo, and A. Jardani (2007b),
A new formulation to compute self-potential signals associated with
ground water flow, Hydrol. Earth Syst. Sci., 11, 1 – 11.
Hansen, P. C. (1998), Rank-Deficient and Discrete Ill-Posed Problems:
Numerical Aspects of Linear Inversion, 247 pp., SIAM, Philadelphia, Pa.
Jardani, A., A. Revil, F. Akoa, M. Schmutz, N. Florsch, and J. P. Dupont
(2006), Least squares inversion of self-potential (SP) data and application
to the shallow flow of ground water in sinkholes, Geophys. Res. Lett., 33,
L19306, doi:10.1029/2006GL027458.
Linde, N., A. Revil, A. Bolève, C. Dagès, J. Castermant, B. Suski, and
M. Voltz (2007), Estimation of the water table throughout a catchment
using self-potential and piezometric data in a Bayesian framework,
J. Hydrol., 334, 88 – 98.
Minsley, B. J., J. Sogade, and F. D. Morgan (2007), Three-dimensional
source inversion of self-potential data, J. Geophys. Res., 112, B02202,
doi:10.1029/2006JB004262.
L24403 JARDANI ET AL.: TOMOGRAPHY OF GROUND WATER FLOW L24403

Revil, A., L. M. Cathles III, S. Losh, and J. A. Nunn (1998), Electrical
conductivity in shaly sands with geophysical applications, J. Geophys.
Res., 103, 23,925 – 23,936.
Revil, A., P. Leroy, and K. Titov (2005), Characterization of
transport properties of argillaceous sediments: Application to
the Callovo-Oxfordian argillite, J. Geophys. Res., 110, B06202,
doi:10.1029/2004JB003442.
Revil, A., N. Linde, A. Cerepi, D. Jougnot, S. Matthai, and S. Finsterle
(2007), Electrokinetic coupling in unsaturated porous media, J. Colloid
Interface Sci., 313(1), 315 – 327, doi:10.1016/j.jcis.2007.03.037.
Richards, L. A. (1931), Capillary conduction of liquids through porous
media, Physics, 1, 318 – 333.
Suski, B., A. Revil, K. Titov, P. Konosavsky, M. Voltz, C. Dagès, and
O. Huttel (2006), Monitoring of an infiltration experiment using the self-
potential method, Water Resour. Res., 42, W08418, doi:10.1029/
2005WR004840.
Tikhonov, A. N., and V. Y. Arsenin (1977), Solutions of Ill-Posed Problems,
John Wiley, Hoboken, N. J.
van Genuchten, M. T. (1980), A closed-form equation for predicting the
hydraulic conductivity of unsaturated soils, Soil Sci. Soc., 44, 892 – 898.
Zhdanov, M. (2002), Geophysical Inverse Theory and Regularization Pro-
blems, Elsevier, Amsterdam.

W. Barrash and B. Malama, Center for Geophysical Investigation of the
Shallow Subsurface and Department of Geosciences, Boise State
University, Boise, ID 83725, USA.
A. Bolève and A. Crespy, UMR 5559, LGIT, CNRS, INSU, Equipe
Volcans, F-73376 Chambéry, France.
J.-P. Dupont, UMR 6143, CNRS, Département de Géologie, University
of Rouen, F-76821 Rouen, France.
A. Jardani and A. Revil, Dept. of Geophysics, Colorado School of Mines,
1500 Illinois St., Golden, CO 80401, USA. (arevil@mines.edu)

6 of 6